Summary
Riley Brown turns Bluey, an open-source iOS and Mac app he vibe coded the week before, into a focus companion for people with ADHD. Bluey runs on an old phone beside the computer, follows the mouse, points at things on screen, and comments with a playful voice. The Mac and iOS apps pair through a QR code and live in a single monorepo on GitHub.
The original version used OpenAI’s Realtime API (GPT Realtime 2.1) with GPT-4o Transcribe for voice input. For the new version, Brown introduces OpenAI’s Decisions API, a specialized inference API powered by GPT-6 Luna that returns a predicate probability, a choice among defined options, or a score instead of text. Because it produces no text output, it is fast and cheap: a hundred screenshot checks cost about a cent, and eight hours of checks every 30 seconds costs roughly ten cents.
The plan is for the Mac app to capture a screenshot every 30 seconds and send it, along with the stated task, elapsed time, idle duration, and recent history, to the API to classify the user as on task, off task, or uncertain. Brown then has Codex build an intervention algorithm so Bluey only nudges after repeated or prolonged distraction, keeping it helpful rather than annoying.
Viewers can follow along with the prompt and repo to build their own version.
📺 Source: Riley Brown · Published November 02, 2026
🏷️ Format: Hands On Build
